How we compute fair value
Most sites show you a valuation they bought from a data vendor. Every fair value on Graham Terminal is computed by us, from reported fundamentals — which means we can show you how it was built, how much data backs it, and when we don't trust it enough to publish it. This page explains each of those.
One number, several models
A published fair value is a composite, not the output of a single spreadsheet. We run a family of intrinsic models — discounted cash flow to the firm and to equity, a dividend discount model, earnings power value, residual income — alongside relative models built on peer multiples. Each model reports its own value and its own confidence, based on how well the company's financials fit that model's assumptions.
- Model weights are routing weight × model confidence, renormalized — a model that fits the business poorly contributes less.
- The relative family is hard-capped, so peer multiples can never dominate an intrinsic view.
- Disagreement between models is surfaced as a range, not averaged away. A composite is never published without its range.
What we refuse to publish
A fair value we wouldn't defend is withheld — shown as diagnostics on the stock page, but never published as a headline number:
- Models below a minimum confidence are excluded from the composite.
- If too few independent model families survive the gate, there is no composite at all — one surviving model is reported as indicative only.
- Implausible outputs are excluded rather than blended in.
The Graham score
Every covered company carries a single 0–100 grade, built from five pillars worth 20 points each:
- Financial strength — can the balance sheet take a punch.
- Profitability — how much the business earns on what it employs.
- Growth — whether revenue and earnings are actually compounding.
- Valuation — what you're paying for all of the above.
- Momentum — whether the market is moving toward or away from it.
Each pillar is a basket of reported metrics, and every metric is judged against sector-specific bars — a good margin for a software company is not a good margin for a grocer — so scores compare fairly across industries. A metric that doesn't apply to a company is dropped and the rest of its pillar rescales; a pillar with too little data is withheld entirely rather than scored on guesswork. Scores are recomputed three times each trading day, with live prices feeding the valuation and momentum legs.
How much data backs a score
Our composite quality scores state their own coverage: how many of the inputs a score wants were actually available for that company. A 78 built on complete statements and a 78 built on half of them are different claims, so we tell you which one you're looking at.
Checked against the filing
For the contested lines of the financial statements, we cross-check our stored figures against what the company actually filed with the SEC (XBRL company facts). Where the two diverge beyond a small threshold, the stock page shows both numbers side by side — the filing's and ours — instead of silently preferring either.
Every number is dated
Every computed figure on the site carries a visible "as of" — the date or age of the data it was computed from, with the method one tap away. A stale number that looks live is the worst failure a research site can have, so freshness is rendered on the page, not hidden in a tooltip.
What this is not
A fair value is a model's estimate under stated assumptions, not a price target and not investment advice. Models can be wrong, inputs can be late, and a wide margin of safety can reflect a risk the model doesn't see. Read the financial disclaimer, and inspect the assumptions yourself — every model is open on the stock page's Valuation tab, where you can change them.